Semantic Information Extraction using Deep Learning

语义信息检索通过开发机器学习和信息论领域的新算法，研究将集中于从海量数据中提取有意义和可操作的信息。主要有两个重点：(1)提取给定多媒体输入的语义属性和属性之间的关系以及语义字幕；(2)给定语义属性或字幕，从数据库中检索相关的多媒体文件。典型的人工智能算法将输入分类到不同的预定义类别中，并抑制与预定义类别无关的视觉信息。提取语义字幕将使人工智能算法能够构建更丰富的底层视觉数据表示。

Semantic Information Retrieval The research will focus on extraction of meaningful and actionable information from bigdata by developing novel algorithms in areas of machine learning and information theory. There will be two main focuses(1) extract semantic attribute and relationship between attributes as well as semantic captions for given multimedia input,(2) given semantic attributes or caption, retrieve relevant multimedia files from databases. A typical AI algorithm classifiesinputs into different predefined classes and suppresses visual information which is irrelevant to predefined classes.Extraction of semantic captions will allow AI algorithms to build a much richer representation of the underlying visual data.